DEEP NEURAL NETWORKS AIRPORT RUNWAY CRACK DETECTION AND DENSIFICATION

Abstract

As a greater number of people travel due to the built many airports by improved Infrastructure, a lot of runways are used constantly by planes and worries about safeties are increasing since new ways of using them keep on mushrooming because of this line of action. Runway maintenance has become an essential duty due to check of cracks. However, traditional methods such as manual inspection because cracks always have differing degree on division have never had good performance in time and needed more time to proof the exact problem. To solve these problems, it is possible to present an dataset that goes by the name of ARID. It splits into eight divisions of runway cracks. It contains 8228 anotated illustrations hence serving as a good source for acquiring the models to work with while training or evaluating them. By using modern methods of deep learning like YOLO v5 and Faster R-CNN, this idea creates a good working algorithm for detecting runway cracks. To get optimal model performance tuning and changing different parameters being fine-tuned are involved. We have noticed a significant improvement in the performance metrics of our model as a result of intense experimentation. The precision of crack detection has increased from 83% up to 92%, while the recall rate has escalated steadily from an initial 62.8% to its current level which stands at 76%. The proposed model has been demonstrated by these results to be effective in precisely identifying and categorizing cracks on runways, so that runway maintenance and safety can be upgraded.